Unreliable minds, unreliable machines: dyslexic memory, ChatGPT, and the epistemic disobedience of generative AI

Abstract This paper examines the epistemic entanglement between dyslexic cognition and generative artificial intelligence (AI), focusing on ChatGPT-4o. People with dyslexia have long relied on digital writing technologies, including speech-to-text software, spell-checkers, and grammar correctors, to scaffold memory and expression within institutions organised around normative expectations of fluency, coherence, and reliable recall. With the emergence of generative AI, this assistive relationship undergoes a significant transformation. Systems such as ChatGPT produce confident approximation, fabricated continuity, and hallucinated citation, introducing forms of epistemic instability that resonate uneasily with cognitive tendencies historically medicalised and sanctioned in dyslexic subjects. Rather than treating these resonances as metaphorical or anecdotal, the paper advances a circumscribed epistemological argument grounded in the autoethnographic method. A series of writing encounters is analysed as situated interactions in which neurodivergent cognitive orientations and synthetic forms of knowing are brought into relation, revealing how cognition and epistemic authority are distributed across hybrid human–machine assemblages and how norms of reliability are differentially governed. Drawing on critical disability studies and critical AI scholarship, and developing the concepts of dyslexic thinking , the dyslexic regime of forgetfulness , and dyslexic remembering , the paper reframes dyslexic cognition as a disobedient epistemic orientation rather than an impairment requiring remediation. It traces a broader shift in technological companionship from corrective assistance toward collaborative yet unstable forms of knowledge production, situating generative AI within the institutional and political economies that organise contemporary epistemic practice. The analysis does not propose equivalence between neurodivergent cognition and machine intelligence. Instead, their analytic friction exposes the asymmetrical conditions under which unreliability is tolerated, optimised, or penalised, advancing unreliability as a site of epistemic inquiry rather than a technical or cognitive failure.